Académique Documents
Professionnel Documents
Culture Documents
December/2008
December/2008
.
.MATLAB-7.5
Introduction:Load-Flow is very important and fundamental tool for analysis any power
system and is used in the operational as well as planning stage. Certain applications,
particularly in system automation and optimization of power system, require
repeated Load-flow solution. In these applications it is very important to solve the
load flow problem as efficiency as possible [1]. Almost all the known methods of
numerical analysis for solving a set of non-linear algebraic equations have been
applied in developing load flow algorithms. One or more desirable features to
compare the different load flow methods can be the speed of solution, memory
storage requirements, accuracy of solution and the reliability of convergence
depending on a given solution. Though, robustness or reliability of convergence of
the method is required for all types of applications, the speed of solution are more
important for on-line applications compared to the off-line studies [2].
Many researches on using Artificial Neural network in power systems had
been studied; Masanori, S. and et..al in [3] use artificial neural network to calculate
various load flow state in power system since the achievement results shown that
ANN network is a fast tool to perform the large scale power system calculations.
Nguyen, T. T. in [4] presents the development of a ANN networks architecture
which implements the Newton-Raphson algorithm for solving of power network
load flow equations. Jayendra Krishna and Laxmi Srivastava in [5] develop
counterpropagation neural network to solve Load-Flow problem under different
loading contingency condition for computing bus voltage magnitudes and angle of
the power system, the composition of the inputs variable of the proposed method has
been selected to emulate the solution process of a conventional load flow program.
Hung-Cheng Chen and et al in [6], proposed partial discharge classification using
neural networks and statistical parameters commercial partial discharge detector is
firstly used to measure the 3-D partial discharge patterns of epoxy resin power
transformers. Then, the gray intensity histograms extracted from the raw 3-D
partial discharge patterns are statistically analyzed for the neural-network-based
(NN-based) classification system. Noor Izzri Abdul Wahab and et..al in [7], presents
transient stability assessment of electrical power system using probabilistic neural
network (PNN) and principle component analysis. Transient stability of a power
system is first determined based on the generator relative rotor angles obtained
from time domain simulation outputs. In this paper we used Radial Basis and Error
Backpropagation neural networks with Decoupled load flow method to raise the
speed computation process and minimize the memory storage which required to the
Load-Flow analysis of Iraqi National Super Grid (400 KV network).
Static Load-Flow Analysis:The objective of Load-Flow analysis is to determine the voltage and its angle
at each bus, real and reactive power flow in each line and line losses in the power
system for specified bus or terminal conditions. The power flow studies are
conducted for the purpose of planning (viz. short, medium and long range
planning), operation and control. For the purpose of power flow studies, it is
311
December/2008
assumed that the three-phase power system is balanced and also mutual coupling
between elements is neglected. Variable associated with each bus of the power
system include four quantities viz. voltage magnitude Vi, its phase angle i , real
power Pi and reactive power Qi total 4n variable for n buses system. At every bus
two variables are specified, the remaining two can be found by solving the 2n power
flow equations. Out of these four quantities only two are generally specified at a few
buses and depending upon which two are specified, we have three categories of
buses, namely Swing Bus or Reference Bus, Generator Bus or PV Bus and Load Bus
th
or PQ Bus. From the nodal current equations, the total current entering the i bus
of m bus system is given by:-[5]
m
(1)
k 1
At i
th
S P j Q V I V (Y V
*
or Pi j Q V i Y ik V k e
i
k 1
P Re al V Y V e
k 1
P V Y V
i
k 1
ik
Or
ik
j(
ik
ik
j(
ik
k 1
(2)
C os ( ik i k )
k 1
ik
k 1
bus will be
j(
ik
Qi V i Y ik V k C os ( ik
k 1
th
Im ajinary V i Y ik V k e
m
Or
(3)
V iV k [Gik Sin
k 1
i k )
( i k ) Bik Cos( i k )
(4)
Equations (3) and (4) are known as static Load-Flow equation, the load flow
equations used in Netown-Raphson method computation of voltage corrections are
given as:
V J 11
V J 21
J
J
P
Q
22
12
(5)
311
Where:
J 11
Pi
J 12
Pi
V k V k
J 21
December/2008
J 22
V k V k
The solution of Equation (5) provides the correction vector i.e. .s for all the
PV and PQ type buses and .Vs for all the PQ type buses, which are used to update
the earlier estimates of s and Vs. This iterative process is continued till the
mismatch vector i.e. .Ps for all the PV and PQ type buses and .Qs for all the PQ
buses become less than a pre-assigned tolerance value . As can be observed form
(5), during each iteration, Jacobian elements are to be calculated and its inverse is
also required, due to this fact, the Newton-Raphson method requires more time per
iteration.
Decoupled Load-Flow Method:In the strictest use of the Newton-Raphson procedure, the jacobian is
calculated and triangularized in each iteration. In practice, however, the jacobian is
often recalculated only every few iterations and this speed up the overall solution
process. The final solution is determined by the allowable power mismatches and
voltage tolerance at the buses. When solving large scale transmission power systems,
an alternative strategy for improving computational efficiency and reducing
computer storage requirements is the decoupled Load-Flow method, which makes
use an approximate version of the Newton-Raphson procedure. The principle
underlying the decoupled approach is based on two observations:- [8]
larger than
Pi
, which is also
j
J
1
11
and Q
J V
1
22
(6)
To get fast computation in load flow solution, the decoupled load flow program has
been developed in this paper and run to generate several training / testing patterns.
311
December/2008
R
(P.U)
X
(P.U)
B
(P.U)
MMDH-MSL4(1+2)
MSL4-BAJp(1+2)
BAJP-BAJG
BAJG-KRK4
BAJP-BGW4
BAJP-BGW4
BAJP-HDTH
HDTH-QIM4
HDTH- BGW4
BGW4-BGN4
0.001436
0.004195
0.000022
0.001799
0.004832
0.004962
0.003446
0.002918
0.004845
0.000932
0.011768
0.034371
0.000197
0.016351
0.043931
0.045113
0.031323
0.02391
0.044049
0.008471
0.364392
1.064256
0.005837
0.484471
1.301651
1.336673
0.928083
0.740352
1.305153
0.250991
311
0.001582
0.000288
0.000152
0.000867
0.001261
0.00122
0.000433
0.003075
0.000809
0.002326
0.003833
0.004321
0.006248
0.004393
0.001183
0.0013
0.001625
0.000823
0.000433
0.014383
0.00262
0.001379
0.00788
0.011465
0.010149
0.00394
0.027954
0.006734
0.019349
0.034849
0.039282
0.05713
0.039932
0.010756
0.01182
0.014775
0.007486
0.00394
December/2008
0.426101
0.077632
0.040859
0.23348
0.339713
0.318967
0.11674
0.82827
0.211651
0.608124
1.032565
1.163898
1.69273
1.18316
0.3187
0.35022
0.457775
0.221806
0.11674
Table (2): Bus-Bar's data of Iraqi National Supper Grid System (400KV)
Bus-Bar Name
V(p.u)
Pg(MW)
PL(MW)
QL(MVAR)
BAJP
BAJG
MMDH
HDTH
QDSG
MSUP
NSRP
HRTP
MSL4
KRK4
DAL4
BGE4
BGN4
BGW4
QIM4
BGS4
AMN4
KUT4
QRNA
KAZG
BAB4
KDS4
1.02
1.01
1.0
1.0
1.01
0.96
0.98
1.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
500
500
500
242
600
650
380
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
200
90
50
100
60
120
100
38
300
70
150
500
300
500
60
100
350
100
70
350
100
200
98
10
20
60
70
70
54
22
180
40
80
360
200
360
40
50
47
60
30
200
50
100
311
December/2008
Figure (1): Single line diagram of Iraqi National Super Grid system. [12]
Proposed Solution Method:Three layers Radial Basis and Error Backpropagation Neural Networks have
been used in this paper to provide the desired power system model. It is assumed
that the neural network architecture would be sufficient enough to recognize any
non-linearity in the simulated power system. The input vector of proposed neural
networks is 2N (N is the number of system buses) and the output vector is 2(N-1),
the size of hidden layer is based on educated trail and error approach; initially it is
identical on output layer size and then increased as learning/identification process
requires. Figure (2) shows the architecture of proposed neural networks, during the
training process of this network, a set of input values are fed to both neural
networks model and numerical decoupled load-flow iterative method, then the
output of three systems are compared to produce the error signals. The network
architecture specification is as the following: Input layer size equal to 44 neurons represent the slack bus phase angle and
voltage magnitude, and real and reactive powers of other buses.
Output layer size of equal to 42 neurons representing all buses (except slack
bus), estimated voltage magnitude and phase angle.
Hidden layer size equal to 56 neurons found by educated trial and error
procedure.
Input
slack
x
x
Hidden
Output
w
w
slack
11
i1
pv & pq buses
PV & Pq Busses
1p
w
w
im
remain buses
ih
mp
mo
remain buses
Basis Neuron
Figure (2): Architecture of proposed Radial Basis Neural Network
Simulation Results:After the implementation of conventional Load-Flow method program, the
training data of radial basis and error backpropagation neural networks are
311
December/2008
obtained then the conventional and neural programs are performed to achieve the
simulation results which are tabulated in Tables (3) and (4). Depending on the
results of Decoupled Load-Flow method which are considered as a main reference in
comparison due to its high accuracy with respect to neural methods, it was found
that the results of proposed neural networks method obtains a good average time
and accuracy that the neural network results not acceded in estimated values the
permitted error level percentage. Table (5) shows the results of solution methods in
the point of view of the execution time and accuracy. Figures (3) and (4) show the
training curve of proposed Error Back-propagation and Radial Basis Neural
Networks and voltage profile of test system calculated by three proposed methods
has been illustrated in Figure (5).
Table (3): Simulation results of Solution Methods from Voltage Magnitude Point of View
Bus Name
BAJG
MMDH
HDTH
QDSG
MSUP
NSRP
HRTP
MSL4
KRK4
DAL4
BGE4
BGN4
BGW4
QIM4
BGS4
AMN4
KUT4
QRNA
KAZG
BAB4
KDS4
Bus
No.
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
DLF
Method
Bus
(P.U)
1.00000
1.0000
1.0000
1.0100
0.9600
0.9800
1.00082
0.9784
0.9882
0.9775
0.9827
0.9973
1.0041
0.9803
0.9885
1.0015
1.0023
0.9585
1.0188
1.0422
0.9667
EBPNN Method
Bus
(P.U)
1.0000
1.0031
1.0046
1.0140
0.9645
0.9833
1.0
0.971
0.977
0.974
0.990
0.993
1.000
0.983
0.984
1.001
1.00
0.964
1.020
1.040
0.967
RBNN Method
Error
%
0.0000
0.3100
0.4600
0.3960
0.4680
0.3367
0.0819
0.7560
1.1330
0.3580
0.7420
0.4310
0.4083
0.2754
0.4552
0.0499
0.2294
0.5738
0.1177
0.2110
0.0310
(P.U)
0.998
0.990
1.009
1.000
0.979
0.988
0.986
0.989
0.980
0.983
0.995
0.981
1.000
1.000
0.995
0.988
0.992
0.975
1.009
1.0392
0.972
Bus
Error
%
0.2000
1.0000
0.9000
0.9900
0.1979
0.8163
0.0819
1.0834
0.8297
0.5626
1.2516
1.6344
0.4038
1.7341
0.6630
1.3479
1.0276
1.7214
0.9619
0.8878
0.2379
Table (4): Simulation results of Solution Methods from Phase Angle Point of View
DLF Method
Bus Name
Bus
No.
BAJG
MMDH
HDTH
QDSG
MSUP
NSRP
2
3
4
5
6
7
(radian)
EBPNN Method
RBNN Method
Error
Error
0.0342
0.0963
0.0231
0.0765
0.0707
0.0734
(radian)
0.0346
0.0971
0.0226
0.0755
0.0694
0.0720
%
1.1695
0.8307
2.1645
1.3071
1.8387
1.9073
(radian)
0.0348
0.0973
0.0235
0.0733
0.0722
0.0752
%
1.7543
1.0384
1.7316
1.5686
2.1216
2.4523
311
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
0.0671
0.0779
0.0542
0.0642
0.0601
0.0587
0.0589
0.0363
0.0888
0.0540
0.0417
0.0547
0.0429
0.0437
0.0218
0.0659
0.0766
0.0531
0.0629
0.0593
0.0593
0.0594
0.0365
0.0893
0.0538
0.0419
0.0538
0.0422
0.0444
0.0215
1.7883
1.6688
2.0295
2.0249
1.3311
1.0221
0.8488
0.5509
0.5630
0.3703
0.4796
1.6453
1.6317
1.6018
1.3761
December/2008
0.0689
0.0753
0.0554
0.0627
0.0611
0.0595
0.0596
0.0366
0.0895
0.0543
0.0422
0.0551
0.0433
0.0446
0.0222
2.6825
1.7971
2.2140
2.3364
1.6638
1.3628
1.1884
0.8264
0.7882
0.5555
1.1990
0.7312
0.9324
2.0594
1.8348
Table (5): The results of solution methods in the point of view of the execution time
and accuracy
Solution Method
DLF
EBPNN
RBNN
Execution Time(s)
4.46
1.48
0.65
99.618 %
99.117 %
98.659 %
98.436 %
Accuracy
313
December/2008
Voltage Magnitude
1.06
1.04
1.02
1
0.98
0.96
0.94
0.92
0.9
1
11
13
Bus No.
15
17
19
21
DLF Method
RBNN method
EBPNN Method
311
December/2008
References:1- Ray, D. Zimmerman and Hsiao, Dong Chiang, "Fast decoupled power Flow for
Unbalanced Radial Distribution systems", IEEE power engineering society for
presentation IEEE/PES winter meeting, New York, 1995.
2- Van Amerongen R., "A general purpose version of the fast decoupled Load
Flow", IEEE Trans. On power system, Vol. 4, No. 2, 1990.
3- Masanori, S. and et..al, "The calculation of electric power flow using neural
network systems", IEEE Japan, Vol. 114c, No.9,2002.
4- Nguyen, T.T., "Neural Network load flow for on-line static security
assessments", Energy system center, No. 10, 2002.
5- Jayendra Krishna and Laxmi Srivastava, "Counterpropagation Neural Network
for Solving Power Flow Problem", International Journal of Intelligent
Technology Volume 1 Number 1, 2006.
6- Hung-Cheng Chen and et..al, " Partial Discharge Classification Using Neural
Networksand Statistical Parameters", Proceedings of the 6th WSEAS
International Conference on Instrumentation, Measurement, Circuits & Systems,
Hangzhou, China, April 15-17, 2007.
7- Noor Izzri Abdul Wahab and et..al, " An Improved Method in Transient
Stability Assessment of a Power System Using Probabilistic Neural
Network", Journal of Applied Sciences Research, 3(11): 1267-1274, 2007.
8- Grainger, John. J and Stevenson, D. William .Jr, Power System analysis,
Neperdazan, Iran, 2003.
9- Rezaul Haque and N. Chowdhury, "An Artificial Neural Network Based
Transmission Losses Allocation for Bilateral Contracts", IEEE
CCECE/CCGEI, Saskatoon, May 2005.
10-Jain, T. Srivastava L. and S. N. Singh, " Fast voltage contingency screening
using radial basis function neural network", IEEE Transaction on power
systems, Vol. 18, No. 4, 2003.
11-Leonardo, paucar, " Artificial neural networks for solving power flow problem
in electric power systems", Electric power systems research, Vol. 62, 2004.
12- Al- sujada, Jwan S. Ra`afat Khadsim, Enhancement of Transient Stability for
Iraqi Power system by Using Fuzzy Genetic Controller, Ph.D Thesis,
university of Technology, Baghdad, 2007.
311